Data?driven approach to construct a fracture toughness master curve for ferritic steels based on tensile properties

The characterization of the fracture toughness (KJc) ferritic steels in ductile-to-brittle transition (DBT) region, which is vital for analyzing structural integrity cracked structures subjected to large temperature changes, has been pursued by researchers and practitioners. This because KJc associated with a dependence (a change approximately 400% corresponding 100°C), specimen thickness (approximately ? 1/(specimen thickness)1/4), scatter ±100% variation around median value). Therefore, amount data required assessment. Constructing master curve (MC) that can predict arbitrary steel, temperature, size expected benefit engineering community, as testing time- material-consuming above-mentioned issues. While ASTM E19211 MC provides one solution, failure predicting some reported.2, 3 Hence, considering data-driven approach reexamining physical background may enable construction applicable all steels. constructed considers tensile properties explanatory parameters avoids tests, thus holding higher merit than E1921 MC. led hypothesis should have uniform relationship (?YSRT–?YS), it was selected main parameter. On basis this selection, realization based on Equation 1 an In addition, (?B/?YS) reactor pressure vessel (RPV) non-RPV steel 1.2 1.7, respectively,2 ?B at RT added parameter represent RT; assumed exhibits small considered range. Finally, width W (mm) effect, used regression were limited those thickness-to-width ratio B/W 0.5. section presents requires only test interest under aforementioned assumptions: (i) direct correlation SED; (ii) fundamentally described ?YS, modification ?B; (iii) ?B/?YS changes minimally from its value RT. First, applying Box–Cox transformation7 531 collected literature8 (i.e., cases = 0.5 downloadable data), function g derived logarithmic. Then, GLM using Bayesian conducted assuming prior distribution coefficients 2 (corresponding noninformative distribution) Gamma shape 0.01 scaling Weibull distribution-specific ? (shape factor). For Markov chain Monte Carlo samples, eight chains 20,000 iterations each considered, 10,000 warmup iterations. simulation confirmed converge checking whether less 1.17 unknowns. 50% confidence 4.04, close fixed 4 study, developed data, prediction 5%, 50%, 95% obtained (x%) cx% KJcnRT × exp[0.00458(?YSRT ? ?YS)]. A case. specifying ?YS be predicted, well ?YSRT ?BRT. If clearly Z–A MC, ?BRT are prediction. does not require tests. applied study categorized “data regression” objective variable variables related size. Another prospective category research “determining hidden law” fracture. example, mechanics-oriented useful simple thickness)1/4, but failed explain existence lower bound along Thus, limitations. Meanwhile, conducting elastic–plastic finite element analysis (EP-FEA) 3D criterion specimens, dependence. Although “black box,” integrating EP-FEA post-processor complicated cracks. Our future aims investigate mechanics box” criterion. support findings openly available Appendix E (https://doi.org/10.1016/j.engfailanal.2020.104713).